Applied Research - Computer Use

Role overview Build and improve training and evaluation systems for multimodal agents that use browsers, desktops, and multiple applications. You’ll turn real workflows into tasks and test environments, then run experiments to make agents more reliable across long sequences of actions.

Responsibilities - Create computer-use environments from real workflows, demonstrations, tools, and measurable outcomes. - Train agents to handle changing interfaces, partial failures, and ambiguous intermediate states. - Design evaluations that measure task completion, efficiency, permissions, and appropriate escalation. - Run post-training experiments against meaningful baselines and investigate whether results come from the model, data, harness, inference, or infrastructure. - Test recovery and generalization across applications, layouts, and permissions; share effective methods as reusable research or systems.

Requirements - Relevant experience in multimodal agents, computer use, browser or desktop automation, reinforcement or imitation learning, human-computer interaction, embodied agents, or long-horizon agent evaluation. - Hands-on experience building and debugging model-training, post-training, agent, evaluation, or related experimental workflows. - Strong experimental judgment, including careful trace review, meaningful baselines, and clear failure analysis. - Strong Python skills and the ability to build artifacts other researchers and engineers can run, inspect, and extend. - Comfort owning ambiguous work and keeping collaborators informed as the approach develops. - Evidence of strong work through research, open-source systems, benchmarks, or shipped products.

Benefits and work setup - Full-time role with flexible work location, including remote work or offices in San Francisco, New York, and London. - Listed cash compensation range of $150,000–$300,000, plus equity incentives. - Visa sponsorship and relocation support, professional development budget, and team off-sites and conference attendance.